Two RX 9070 XTs at 4K: What Crossfire Actually Delivers in 2025
AMD’s Radeon RX 9070 XT arrived earlier this year as a strong mid-to-high-end competitor, punching well above its price point in rasterization and holding its own in ray tracing. Running one card is already a solid 4K proposition. Running two raises a different question entirely – not whether you can do it, but whether the scaling makes any financial sense when the GPU market offers so many alternatives.
The Setup and Why Anyone Is Still Testing Multi-GPU
Multi-GPU configurations have been declared dead so many times that most reviews stopped covering them entirely after 2020. But AMD has quietly kept mGPU support alive through its driver stack, and with the RDNA 4 architecture’s efficiency improvements, there is renewed curiosity about whether two RX 9070 XTs running together can carve out a practical use case – particularly at 4K where raw pixel throughput matters most.
The test platform uses a Ryzen 9 9950X paired with 64GB of DDR5-6000, running on a high-end X670E motherboard with both GPUs seated in full-bandwidth PCIe 5.0 x16 slots. That last detail is important. Previous multi-GPU tests were often bottlenecked by bandwidth limitations from PCIe 3.0 or x8 configurations. Here, both cards operate at their maximum rated bandwidth, giving the setup the best possible conditions to show what two RX 9070 XTs can actually do. Pairing a capable CPU like this also matters – as covered in the Ryzen 9 9950X3D streaming benchmark analysis, higher-core-count Ryzen processors handle the thread overhead of complex rendering pipelines far better than mid-tier alternatives.
Game selection for this kind of test requires care. Titles need to actually support mGPU at the driver or API level, and the list in 2025 is shorter than most enthusiasts would hope. The tested titles include Cyberpunk 2077 with its Phantom Liberty expansion enabled, Assassin’s Creed Shadows, Black Myth: Wukong, and Microsoft Flight Simulator 2024 – all at native 4K with FSR 3 disabled to isolate raw rasterization performance. Ray tracing presets are tested separately, since the scaling behavior there diverges significantly from standard rasterized workloads.
Baseline single-card performance for the RX 9070 XT sits between 68 and 84 frames per second depending on the title at 4K ultra settings, which is genuinely competitive against the RTX 4080 in most rasterized workloads. That means the dual-card ceiling, if scaling holds anywhere near 80 percent, would push into territory that no single consumer GPU can match outside of the RTX 4090 – and even then, only in some titles.
The Scaling Numbers and Where Things Break Down
Cyberpunk 2077 returns the best scaling result of the entire test suite. At 4K ultra with path tracing disabled, two RX 9070 XTs average 143 frames per second against the single card’s 79, which works out to roughly 81 percent scaling efficiency. That number is better than most dual-GPU results from the RDNA 2 and RDNA 3 era, and it holds relatively steady across the benchmark sequence with minimal frame time variance. CD Projekt Red’s DX12 implementation has historically played better with multi-GPU configurations than most titles, and that pattern continues here.
Black Myth: Wukong tells a more complicated story. Average frame rates climb from 71 to 118 fps – about 66 percent scaling – but the frame time graphs show irregular spiking that creates a perceptible judder even when the average looks fine. The Game Science engine’s handling of mGPU appears inconsistent between scene types: open outdoor areas scale well, but dense interior environments with heavy particle and lighting effects cause synchronization hitches between the two cards. The average number flatters the actual experience.
Assassin’s Creed Shadows performs somewhere in between. Ubisoft’s AnvilNext engine shows roughly 70 percent scaling at 4K ultra, producing a smooth 128 fps average against the single card’s 74. Frame pacing here is more consistent than Wukong but not as clean as Cyberpunk. Interestingly, enabling the game’s highest ray tracing preset actually improves scaling slightly – up to about 73 percent – possibly because the RT workload distributes more evenly between the two cards than the shadow and foliage rendering does.
Microsoft Flight Simulator 2024 produces the weakest scaling result: 52 percent, bringing the dual-card average from 61 to 93 fps. The simulator’s CPU-heavy simulation model and the way it manages asset streaming means that the second GPU spends measurable time waiting rather than rendering. For a title this demanding at 4K ultra with photogrammetry fully enabled, even 93 fps is a meaningful improvement over a single card, but the efficiency loss is real and worth acknowledging.
Ray tracing results deserve a separate paragraph because the pattern inverts slightly. With full ray tracing enabled in Cyberpunk at 4K, single-card performance drops to 47 fps – barely playable by most standards. Dual-card scaling here hits only 68 percent, bringing the average to 79 fps. The RT workload is harder to distribute cleanly, and the driver overhead increases. Still, 79 fps with full ray tracing at native 4K is a result that no single GPU under $1,500 can match, which contextualizes the trade-off in a different way.
The Real Question: Does the Math Work?
Two RX 9070 XTs at current street pricing cost somewhere in the range of $1,100 to $1,200 total, depending on board partner models and availability. A single RTX 4090 trades in the $1,400 to $1,600 range used, and a new RTX 5080 sits at $999 MSRP. The dual RX 9070 XT setup beats the RTX 4090 in Cyberpunk rasterization, ties it approximately in Assassin’s Creed, and falls behind in MSFS 2024. It beats the RTX 5080 in every rasterized title tested – but the RTX 5080 has no frame pacing issues, works with DLSS 4 Multi Frame Generation, and runs on a single PCIe slot without requiring mGPU driver overhead.
Power consumption is the number that complicates the value argument most. Two RX 9070 XTs under full load draw approximately 600 watts combined from the wall, versus 480 watts for a single RTX 4090 under similar conditions. Over hundreds of gaming hours, that gap adds up. The dual-card configuration also requires a case with adequate spacing, a PSU rated at 1000W or above, and a motherboard that supports dual full-bandwidth PCIe slots – none of which are standard for a mid-range build. So while the raw performance math occasionally favors two RX 9070 XTs over more expensive single cards, the hidden costs in hardware, power, and the fragile state of mGPU game support make this a configuration that only a very specific type of enthusiast should seriously consider. The 52 percent scaling in Flight Simulator – arguably the most demanding 4K title in the test suite – is a reminder that the worst-case scenario for mGPU can wipe out most of the price-per-performance advantage entirely.
